| 作成者 |
|
|
|
| 本文言語 |
|
| 出版者 |
|
| 発行日 |
|
| 収録物名 |
|
| 巻 |
|
| 号 |
|
| 開始ページ |
|
| 終了ページ |
|
| 会議情報 |
|
| 出版タイプ |
|
| アクセス権 |
|
| 権利関係 |
|
|
|
| 関連DOI |
|
| 関連URI |
|
| 関連HDL |
|
| 概要 |
On a real estate portal site in Japan, users can search for text information but not image information. This is because textual data entered by property management company is in an available format fo...r searching, but image data is not. The floor plans which each property management company made in their respective formats are not compatible unlike the character provided in each fixed entry field. Therefore, the user needs to infer what kind of structure from the input character strings (number of rooms, facilities, area, etc.). In this study, we consider a system to search for similar floor plan images by giving a photograph of a construction toy structure as a search query in order to search for floor plan images. Deep CNN model (VGG16) is fine-tuned separately for the floor plan image and the query image. Subsequently, each feature vector is extracted from each learning layer of VGG16. Siamese Network is constructed to calculate similarity between the floor plan image and the query image by CNN learning the match
mismatch between the floor plan image and the query image. This paper gives an overview of the experimental procedures and results performed. This experiment deals with preprocessing of floor plan images and learning of floor plan images. In another experiment, the character information was extracted from the floor plan images.続きを見る
|
Mendeley出力
